论文部分内容阅读
针对基于像素的谱聚类计算量大、效率低,且受到影像斑点噪声影响严重的问题,该文结合极化合成孔径雷达影像的统计特性,提出了一种基于超像素的极化合成孔径雷达影像谱聚类算法。该方法首先用基于梯度分割影像的分水岭算法得到影像的初始分割;然后按区域邻接关系合并含像素个数较少的极小区域得到超像素图像;最后以超像素为基本数据单元,采用修正Wishart距离作为超像素之间的距离度量标准,通过Nystrm逼近的采样方法获得最终的分类结果。最后利用模拟数据和1991年获取的荷兰Flevoland地区L波段稻田数据验证了该算法的有效性,总体分类精度达到了98.17%。
Aiming at the problem that pixel-based spectral clustering is computationally expensive, inefficient and seriously affected by image speckle noise, this paper presents a super-pixel based Polarimetric Synthetic Aperture Radar Image spectral clustering algorithm. In this method, the initial segmentation of the image is obtained by using the watershed algorithm based on the gradient segmentation image. Then, the superpixel image is obtained by incorporating the minimum area with fewer pixels according to the regional adjacency relationship. Finally, using the super pixel as the basic data unit, Distance As a measure of distance between superpixels, the final classification result is obtained by Nystrm approximation sampling method. At last, the validity of the proposed algorithm is verified by simulation data and L-wave field data in Flevoland, Netherlands obtained in 1991, and the overall classification accuracy reaches 98.17%.